Hacker News Reader

Comment by zallarak | Hacker News Reader

The hft stuff is actually what I'm most skeptical of. I think the barrier to entry is huge in terms of knowledge and initial costs, unless you are an outlier. The opportunities are in more basic things. Mixing a decent quant background with a solid financial background is a potent combination for succesful fundamental value investing. Creating an automated tool to help structure the best ways to get equity exposure using options based on live data is another immediately foreseeable benefit.

View on HN

What are those initial costs? Say I have the knowledge or ability to acquire it, how would I actually go about connecting myself to the great stock feed via (e.g.) Python?

I don't know. It's something I know very little about. I speculate that you have to buy a seat or subscribe to a service at a major stock exchange (very expensive) or work through a third party/broker (still pretty expensive, but also less efficient). The reason behind my skepticism is that the only people I know making sustainable (and great) money from it have very deep pockets to begin with or are huge established players.

I was once visiting at a quant fund where they purchased an office building to get physically closer to a stock exchange. It was very expensive real estate and they were a massive fund. They had a lot talented people. Hard to compete with that.

I also read in the Economist a while ago (not sure if its really true, but its believable) that hedge funds are the largest consumers of computer processors in the US. But again, this is something I know very little about. I could be wrong.

if you want to play in HFT you want 1) wicked chops 2) collocation with exchange computers to pick up nickels faster than anyone else (was going to say sub-millisecond but my knowledge is not that granular) 3) access to historical tick data to analyze, which is expensive to subscribe to and manage (was going to say petabytes but my knowledge is not that granular)

The way to think of a big part of HFT is really just automating the old function of market-making, adjusting the bid-ask on many securities to keep things in line.

In the old days if you wanted to trade, you called a broker who took it off for a price, the bid-ask, and you tried not go get ripped off too much. Now institutions who want to trade look for ways to feed them into the maw of HFT without getting ripped off too much, and HFTs try to figure out how to get as much of a bid-ask as possible.

Surely the biggest consumers of processing power are the data centers of Google, Amazon etc... Places like Goldman have sizable clouds but I don't think they compete yet.

"was going to say sub-millisecond but my knowledge is not that granular" <-- round-trip times are, to 90th percentile, sub-100-microsecond at colo

"was going to say petabytes but my knowledge is not that granular" <-- entire market data is sub-100GB per day.

" have sizable clouds but I don't think they compete yet." <-- cloud computing is not used in HFT

"work through a third party/broker (still pretty expensive, but also less efficient)" <-- there are a glut of brokers and a lack of market participants, so the costs have naturally fallen. 250K will get you time and tick on 4:1 leverage

If you are really interested, it would make sense to start now before rule 15c3-5 kicks in

There are plenty of "retail" ways to get access to this data, both realtime and historical. For realtime data and trading API access, interactive brokers provide a good service. For pure data feeds, both realtime and historical, check out esignal, cqg or iqfeed.

IB data isnt the actual exchange data. For example, on nasdaq, the raw data is the ITCH tick-by-tick data.

iirc iqfeed gives you a normalized data format.

Yeah, the quality isn't the same as a proper feed but i guess the cost doesnt compare either. CQG seem to do pretty good feeds but i've never tried them

$350K soup-to-nuts.

Reply on news.ycombinator.com